Problem
We study how the frequency of priority gas auctions (PGA) affects searcher bidding and competition. PGA on Base provides a natural experiment to estimate the effects of auction cadence causally. Before July 2025, transactions within each 2-second block were ordered through a single PGA round. With the introduction of Flashblocks, each block has ten rounds of PGA. So, the fee mechanism remained unchanged, but searchers effectively got 10× more opportunities to compete.
Arbitrum One recently introduced PGA with a 125 ms round length. The round length can be set to 250/K ms for K between 1 and 10. This makes PGA frequency a tunable parameter for the chain operator.
The natural question is: what happens to searcher competition when the auction becomes faster?
Methodology
A simple comparison of fees before and after Flashblocks is not enough, since market conditions also changed over this period.
We therefore compare 3,032 active searchers against 8,053 similarly active addresses that do not exhibit competitive searcher behavior. Using address and day fixed effects, we estimate how searcher fees changed relative to this control group after Flashblocks were introduced.
Flashblock boundaries are not observable on-chain. We reconstruct them heuristically by following the sawtooth pattern observed in priority-fee variation.
We also match searcher transactions to the liquidity pools they compete over. This lets us study individual arbitrage opportunities and check whether latency matters.
Finally, we develop an auction model in which the winner pays its full bid and losing searchers pay a fraction of their bids (because of early reverts). Searchers can arrive in the current round or a later round. A searcher arriving in the next round loses to a searcher in the current round, regardless of its bid. We derive equilibrium bidding under this setup to study how auction cadence and arrival timing affect competition.
Findings
The headline result of this analysis is that shorter auctions substantially reduce searcher bids. After the PGA round length changed from 2 seconds to 200 ms, the effective priority fee paid by searchers fell by 0.187 gwei relative to the control group. This drop is equivalent to 59% of the fee searchers were paying before the introduction of Flashblocks.
The location of those fees within the block changed as well. Before Flashblocks, 98% of searcher priority-fee value was concentrated in the first 10% of block gas. After Flashblocks, this fell to 28% as opposed to 10% one would expect under shorter auctions. Moreover, competition is spread across the block in ten auctions.
We also find evidence that, under short auctions, some contested opportunities are resolved by latency. That is, a searcher loses an opportunity even when submitting a higher fee because the winning transaction landed earlier. This evidence is suggestive, since some of these reverts are also due to slippage or stale opportunities.
The theory suggests that fees drop in shorter auctions for two reasons. First, during shorter auctions, there is less price movement, which reduces the value of arbitrage opportunities. Second, fewer searchers arrive in time to compete in the same round. However, the model does not suggest that every searcher’s bid is reduced in shorter auctions.
Finally, model-based estimates suggest diminishing reductions in priority fees as auction duration falls further below the observed 200 ms. At 200 ms, most auction windows already have very little competition. Therefore, reducing the auction time even further would not reduce the priority fee by much. This suggests that much of the effect comes from moving from relatively long auctions (2 seconds) to short ones (200 ms).
For details, check out the paper: Priority Gas Auction Cadence and Searcher Competition: Evidence from Flashblocks on Base. Joint work with @edfelten and @kakia